Automated Sleep Staging Using Convolution Neural Network Based on Single-Channel EEG Signal

Sleep disorder diseases have one of the major health issues across the world. To handle this issue, the primary step taken by most of the sleep experts is the sleep staging classification. In this paper, we proposed an automated deep one-dimensional convolution neural network (1D-CNN) for multi-class sleep stages through polysomnographic signals. The proposed 1D-CNN model comprises eleven layers with learnable parameters: nine convolution layers and two-fully connected layers. The main objective of designing such a 1D-CNN model is to achieve higher classification accuracy for multiple sleep stage classifications with reduced learnable parameters. The proposed network architecture is tested on two different subgroups subject sleep recordings of ISRUC-Sleep datasets, namely ISRUC-Sleep subgroup-I (SG-I) and ISRUC-Sleep subgroup-III (SG-III). The proposed deep 1D-CNN model achieved the highest classification accuracy of 98.44, 99.03, 99.50, and 99.03% using the ISRUC-Sleep SG-I dataset and 98.51, 98.88, 98.76, and 98.67% using SG-III dataset for two to five sleep stage classification, respectively, with single channel of EEG signals. It has been observed that the obtained results from the proposed 1D-CNN model give the best classification accuracy performance on multiple sleep stage classifications incomparable to the existing literature works. The developed 1D-CNN deep learning architecture is ready for clinical usage with high PSG data.